Loading summary
Transcript92 lines
- [00:00]
Zach Leach
Let's just come up with this new medium that is about. You write it like a doc and you present it like a deck.
- [00:06]
Rid
You're kind of operating at a whole different level in terms of just how wide the spectrum of possibilities is.
- [00:13]
Zach Leach
I was doing user research and people were like, oh, it made a webpage for me. And we're like, but it said presentation. Like, what do you mean? And it turns out, like, people were like, this would be really cool webpage. Like, maybe I can make this my website.
- [00:25]
Rid
How have you iterated on that flow? There any big sweeping changes where you've learned something or maybe shifted the way that you even think about how to leverage AI effectively?
- [00:35]
Zach Leach
Ultimately, what I ask myself when I'm kind of doing this stuff is like, is this fun to do? Like, is this a fun thing or does it feel like work?
- [00:42]
Rid
When does an actual professional designer working at an AI company take advantage of these tools versus work a little bit more traditionally?
- [00:51]
Zach Leach
Ultimately, what I ask myself when I'm kind of doing this stuff is like, is this fun to do? Like, is this a fun thing? Or does it feel like work?
- [00:58]
John
Welcome to Dive Club. My name is Rid and this is where designers never stop learning. This week's episode is with Zach Leach, who's the head of design at Gamma, which is one of the hottest AI startups right now.
- [01:10]
Rid
So we talk a lot about what.
- [01:12]
John
It'S like as a designer to totally reimagine a category like presentations.
- [01:17]
Rid
From there, we go really deep into.
- [01:19]
John
What it's like working with AI as a new material, and Zach shares a ton of practical stories and lessons learned.
- [01:27]
Rid
So let's start all the way back.
- [01:29]
John
At the beginning because it was a bit of an unorthodox journey.
- [01:32]
Zach Leach
We worked at optimizing together for a long time. We were riding the wave, man. We were like, this thing is, you know, it's a big company, you know, a lot of. A lot of interest and the number one, a B testing platform. And there was like probably three or 400 people at the time, but there was sort of a core group, you know, in the product area who worked really, really well together. You know, so many cool things got made out of that and so many cool moments to work together. It really was about the people when this whole thing started, right? I mean, you got to think about the context. Like, optimizely was, you know, this was like 2021 or 2020 or something. So it was like sort of COVID started to happen and then, you know, there's a lot of uncertainty. In the market, and a lot of our customers at Optimizer kind of like, I don't know, you know, and then we started doing the remote thing, and that was weird. And really, when it all kind of felt like. I say the catalyst really was like the sort of the acquisition, and it was, you know, it was a bumpy landing. I think after that a little bit, you know, we're kind of, like, getting our footing again, trying to see what it's like working for, you know, you know, these new owners and a lot of leadership changes and stuff. And we started just thinking about next steps. I think when the wind kind of falls out of your sails, you're sort of like, what do we do now? You know, so it was a little bit of a lull, and then, you know, conversations started happening. And really, for me, it was. It was John and the potential of working with somebody like Jess, and just. Just. It really came down to the people and. And we had nothing. I mean, there's no. We had just an idea, you know, about just, like, how we could help people share ideas, you know, and make this new medium. And then the funniest thing is that's totally not what we did at first, right? Because you think about it, like, everybody's remote and zoom was, like, blowing up, right? Everybody was doing, like. There was, like, butter and, like, meet or something. Meetup or meet town. There's all these, like, different toys and things that, you know, people were coming up with to get while everybody's remote, to work together. So we came up with this idea of, like, this sharing ideas in kind of a shared meeting space almost. We called it the Lobby. It didn't really have a product name. There was no product, but there was just a tremendous amount of prototyping this thing. And it still works, actually, internally, we can still get to it. But for six months, we built this, like, video chat app, and it was the weirdest time. And it had, like, sound effects. There was, like, a fart sound effect. I remember an investor was like, oh, I think something's wrong with the trumpet, because he kept putting the fart sound. It didn't sound like it's Trump. Anyway, so through that, we sort of evolved. We're like, okay, so we're going to meet in these lobbies, right? We're going to have, like, a meeting. Well, maybe we need an agenda, and maybe the agenda is like an outline. So we built an outliner, and then there was, like, this shared collaborative outliner. And so everybody's in the lobby. You could see Each other, you can see each other's cursors and then you're all typing in this agenda and then you go one by one and then we're like, well, wait a second, maybe we're going to share like a Google Doc. We're going to share a webpage and it's going to be part of the outline. And so you can start to see where we're like, maybe this is actually a presentation tool, right? So you've got different slides, which could be web pages or Google Docs or whatever they are, and then this outline and then the shared meeting space. And eventually we totally dumped all the video stuff, like, let's just build this, this presentation tool. And at the time there was no, there's no AI or anything. And so we kind of set out to say, let's just come up with this new medium that is about. You write it like a doc and you present it like a deck. So it just looks beautiful by default, you know, you type it out just like a doc. And it's important to, to kind of understand like this medium because when AI happens it actually makes a lot of sense because imagine a doc and it's all, it's all linear, right? So we're building this, we're building this like new medium. It's going to be the sort of a linear thing. And then LLM started being a thing, right? And you're like, okay, if you know how they work, they're sort of these prediction engines. They were at the beginning and so they're really good at sort of linearly predicting what's going to come next. And so they work really well for our medium. It was this moment where we realized what we had been building for at that point, eight months. This doc that turns into a deck actually just works really well for LLMs. And you can just pipe it in and say, keep going, keep writing. This slide deck, basically it's like a node based, kind of HTML based sort of format. It actually worked really well now back then, of course, LLMs were not as good as, as they were now. And, and there was a lot of trepidation, I think, about trusting it, about bias, about quality, about all sorts of stuff. And it was a big leap of faith, I think, to say, like, yeah, we're gonna let this thing write a deck on behalf of customers and we're gonna prompt it and it's gonna make.
- [06:03]
John
Good stuff, real quick message and then we can jump back into it. So I'm working on a lot of video and playback features for the first time. And honestly, I have no idea what I would do if it wasn' you can use natural language to search for anything that you're working on. They even have visual search so you can upload a screenshot of your designs and then see relevant UI for inspiration. But it's a total cheat code and I cannot imagine designing without it. So head to Dive, Dot, Club, Slash Mobin to get started today. If you're working at a startup right now, then I have some pretty good news for you because Framer just released an incredible offer that gives startups for free access to their launch plan for a year. That's $900 to design and ship your site with Framer, which is a total no brainer. I mean, I even took advantage of the deal myself, so definitely don't miss out on this offer. You can head to framer.com startups to learn more.
- [07:00]
Rid
Okay, now onto the episode. I want to go into all of those challenges, but maybe first I want to double click on what it's like trying to design a new medium. You know, because I would imagine it's tricky where, you know, on one side you have this feature, rich Google Doc and everything that people expect out of a doc. And on the other side you have slides, which is pretty cemented mental model for people where that is very much of the language of business. So how do you as a designer figure out the right middle ground and where to pull from familiarity versus where to create something new? What was that process like for you?
- [07:36]
Zach Leach
Yeah, that middle ground is, is the trick, right? Because if you want to say this thing does slides, it kind of needs to still feel like slides. But if you also want to say we're leaving slides behind and it's a new way to make slides and presentations, you know, you can't take too far believe, right. I think that we started maybe a little too unstructured, maybe a little too, a little too far from slides. We have a few different features that are interesting and that feel very not like slides. Like, you can put, they're called nested cards. You know, the document is made of cards and you can sort of put a card inside a card. And for a slide tool, you're like, why would I put a slide inside a slide? You know, as the product evolved and as we sort of found our way, we realized giving people little hooks and little anchors, little affordances to sort of still ground in familiarity, but also push a little bit on what the medium could be. So a lot of it is just learning and talking to People and giving people these little, these little ledges to kind of hang onto. Something like export, you know, is like, sure, we could make a new medium and we could say like this is the future, but we sort of still have to play nice with like the rest of the world, right? You know, some CEO somewhere at maybe not a super tech forward company isn't going to necessarily want to have a Gamma link. They're going to want like a PDF or whatever, right? It's kind of an older, older school kind of kind of vibe, right? So we still have to operate in sort of reality while still finding the right things to push on. The sort of guiding light here was writing it like a doc and being able to just have that familiarity of you're not dragging an image in and resizing text boxes and making sure things fit right and flow right. You're just going to type it out, right? And you're just going to kind of say, you know, put a heading in, put bullets in and build out stuff, you know, in a very sort of traditional way. And then when you press the present button, it's going to look like a great slide deck, right? You know, when you say like there's a new medium for kind of sharing ideas, people make all sorts of interesting things, right? It is sort of presentation focused, but people make, you know, web pages, people make all these sorts of different artifacts. People do note taking, people make, you know, leave behinds for sales. There's just so many different little. Education has been a big thing too, like learning courses and stuff like that. And so you get to see people make and it's, it's super fun and interesting. And so much of what we do at Gamma is listening to people, listening to customers and understanding their needs and trying to deliver kind of things that make sense and work with a broader picture, but still can speak to kind of some familiarity.
- [10:01]
Rid
I feel like a lot of product teams right now are dealing with this kind of open ended nature of their product as soon as you incorporate AI because it can kind of do everything and you have that element, but then you also have a product to begin with that was very open ended. And so you're kind of operating at a whole different level in terms of just how wide the spectrum of possibilities is.
- [10:22]
Zach Leach
It's funny because our original investors were like, no, you should like focus on a particular vertical, right? You guys should focus on salespeople or focus on education or whatever it is, but you should build. I think that's very common sort of advice, right? It's like Build one use case out and build it really well. And we're like, no, we're going to build this crazy broad thing that can kind of do anything. And I think it served us well ultimately, sort of, you know, going maybe against convention here. I think the trade off is we've been able to find our sort of customer and find people that have gotten a tremendous amount of value out of it. But the trade off is like, is it bespoke enough, is it sort of sharp enough to work in these specific cases? The future has revealed itself, so to speak, and we can start to build some of those really, really cool, like, specific cases. Webpages, you know, is an interesting sort of concept for us too. Like we did, we originally were going to do this sort of new medium and presentations were kind of the core of it, and maybe some documents. And I was doing user research and people were like, oh, it made a webpage for me. And we're like, but it said presentation, like, what do you mean? And it turns out, like, people were like, this would be really cool webpage. Like, maybe I can make this my website. Maybe we can, I can host this or whatever. And so we started this whole process of building out an entire webpage product which, which is. We're still working on it, but. But I think people have really found it very useful and found like, spin up a website really quick. And a website that you just type like a doc and not have to like fiddle with anything that just looks good by default. Like, that's, that's huge. AI can, can play into that. You can just basically have AI write your web page linearly, like it, like a dog.
- [11:54]
Rid
So that brings up something that I wanted to ask you about specifically the role that AI plays in kind of the customization flow. Because I could see an obvious spectrum that I'm sure you're probably wrestling with as a designer, where on one hand you have just power and customization and the other you have these guardrails to make sure that people can't make a really, really crappy webpage, you know. So how do you think about that tension and how do you think about the design strategy for where you all want to fit onto that spectrum and how does that ultimately take shape in the interface?
- [12:24]
Zach Leach
It starts with principles. One of the core principles we have here when we're doing sort of anything is it's beautiful by default, right? So we actually think even if you don't really use AI, it should still be beautiful by default, but it definitely applies to, to AI. And when you use AI generation tools within Gamma. So what it really means is like, you sure, you can say for AI image editing or AI image generation, you can make an AI image and on one hand we could say, make it whatever style you want or make it look, you know, whatever kind of tone or vibe or theme of the image that you want. But actually what we've done is we've sort of said, well, because the deck has a theme, we're going to automatically sort of apply the colors and the vibe of your theme of the deck to the image, right? So by default, that image you make is informed by the theme of the entire deck. And so you'll get the same color tones, the same illustration style, the same photography style, like whatever it happens to be for that deck. Images are just one example of us saying, give it sort of the ability to be beautiful by default. That's not to say you couldn't break out of it and like, and like go in and say like, no, I want it to be like, kind of crazy. Ultimately, people that use Gamma are going to kind of make whatever they want to make. But we really, really strived for like logical defaults, defaults that just like work and look good and you don't have to, like, think about it too much. A lot of our customers aren't designers, right? They're people who are like, I just want to make a presentation and have it kind of look good. And I think that's really what we try to deliver with AI by, you know, giving it the ability to just make stuff that looks good by default. And if you want to kind of break out of that, have at it. But, but it's, it's easier to make something looks good. It's harder to make something looks bad in Gamma. And that's kind of the guiding principle.
- [14:09]
Rid
Are there other guiding principles that you're kind of turning to on a regular basis as a designer?
- [14:14]
Zach Leach
Obviously we all know like, AI is non deterministic, right? You're going to get what you're going to get out of it. And so being forgiving, I think is really important. Being clear, like what's going to happen before it's going to happen. So a great example is like, if you want to edit your deck with AI and it's going to do some massive manipulation, it's going to like move all these cards around and like really reshuffle stuff. Like, should we let you know it goes and does that? Right? Like, how much of that do we want to bubble up before? And then how much of that do we want to let you undo and really sort of work with it a little bit. We had a principle around suggestions and being able to like giving you a suggestion that's just going to work nice and work well and be beautiful. I've really found that giving just LLMs and giving the models just more context is always going to turn out usually better. Say you're again doing like an image in gamma. It's cool if it knows like what the theme is and maybe the name of the deck, but if it knows like what the card is or if it knows what paragraph it's connected to, you actually can say, you know, let's say I have like a deck about dogs or whatever, right? And it's like different breeds and it knows, like, here's a paragraph for each breed. Like it'll know I'm not just going to make a dog. Going to make a dog of this specific breed. And the specific part of the card and so kind of giving it more context and letting it kind of run with stuff has actually been really cool. That's not really principle.
- [15:28]
Rid
But are you thinking of that as like a core part of the design deliverable now?
- [15:32]
Zach Leach
Yeah, I mean, I think you have to. We have this AI image chat which lets you basically say, here's an image. What do you want to do with it? Right? Like change the lighting or add whatever to this image or put a hat on, whatever the shark, whatever it might be. And I was trying to think, what can we tell the AI? Like, what can. What context can we give it to help you basically predict or know better what it should do, right. And so I was thinking like, maybe a context would be cool if it knew like what you've recently written or where your cursor was or like what you recently deleted or something, right? Like, how much of this can we sort of feed into it? And you kind of have to design that. We have just kind of say like. Like here's this interesting kind of data that we can pipe in and then it's about prototyping it a little bit, right? Like, what's actually going to come out of that? How is it actually going to work if you do feed in all this stuff and it kind of. What's it going to spit out?
- [16:21]
Rid
It's only partially relevant, but it reminds me of this tweet that I saw. It might have been from Jordan Singer. I'll have to double check and I'll put it in the show notes, but it was basically like more AI products should have your clipboard as part of the context that you're feeding the AI. And that kind of was an eye.
- [16:36]
John
Opening moment for me.
- [16:37]
Rid
It's like, oh wow, you know, I wouldn't have thought of that. But you're right. Like that is part of what it looks like to design these products now.
- [16:43]
Zach Leach
The more it knows, the better. Right. And I think where we're headed is a future in Gamma where not only does it know what this deck is about or what this card is about, but it knows like about your business. It knows about like what you're doing, it knows about you do this quarterly business report, whatever, like with, with three different customers every quarter. Like I already know what the format is. I already know like kind of about them. And then maybe it knows those previous stacks and it knows, oh, the last two quarters. Here's the data for that. Right. Let me do a quick summary card for you. Right. So as the context windows get bigger and as you can sort of experiment and add more context, things are only going to just kind of be smoother and easier and smarter, you know, in something like Gamma, where it's going to kind of know a little bit more about you, about your business, about kind of what, what things should look like and how things should feel.
- [17:28]
Rid
It's cool to think about. Like it's almost like a zooming out from just pure solving problems as a designer, but like a way that you can create real value for a company, especially in an early stages, is actively thinking about, okay, how can we design a system where context can become a long term moat for our product?
- [17:45]
Zach Leach
Just recently we're thinking about how, you know, all your images can live in sort of a common place. Right. And you can maybe reuse those. And I thought ChatGPT just released like the library thing for their IM generator and smart. Right. It's like you're using this for work and you want to just kind of capture everything and hold on to everything. And again, they're building state. I mean they get it too.
- [18:04]
Rid
I want to drill into the specific product applications and how you're thinking about AI and maybe we could even look at that creation flow. Like you said earlier on, it does feel like a pretty obvious use case for AI. I'm sure it didn't take you long to imagine how that could play a really big role. But how have you iterated on that flow? There any big sweeping changes where you've learned something or maybe shifted the way that you even think about how to leverage AI effectively where you've made some changes?
- [18:33]
Zach Leach
Yeah. So the first thing we ever did was just a single prompt. And this is like in a very prototype world. Again, I was like, there was no chatgpt at this point. It was like there were very, very, very basic models. There was like, make a deck on this topic, right? And it would take like a minute or like, just like really slow. First thing you see is you need more guidance, right? But the real breakthrough is when we realized, like, you know, maybe we can provide it an outline, or maybe it can make an outline, right? And so one of the really cool things in Gamma, and I think one of the real breakthroughs we had in AI is like the ability to, say, start with an outline or start with a topic and then generate an outline. And it's almost like you get this intermediate step to sort of really hone in what you want to say. And before the deck gets generated, before you have to, like, take in all of this information of all the different cards and all the content on the cards, start with this sort of skeleton. And we're always trying to figure out how detailed, right, the outline should be, should it be able to have images, should it be able to have layouts, right? Like, what is this sort of proto deck, so to speak? And where we've kind of landed right now is you can start with a topic or start with a prompt. It'll generate an outline. You can tweak the outline and sort of, say, add in a few kind of bullet points or cards or whatever, and each of the sort of steps, the outline becomes an individual slide. And you can even say things in the outline, use images about this or whatever. And from that outline, then you get sort of the full deck generation. So early on, it was a single prompt. We realized an outline or something like that would be really useful, so you don't have to wait for the whole thing and you can sort of tweak it and then. So that worked pretty well. We also started with a chat interface too. It was sort of like a back and forth, like, what's the deck about? Would you like me to add these? How would you like this? It was a lot of questions. We iterated away from that and met with something a little more straightforward, a little more like a setup page versus like a. Like a chat interface. And that was really around kind of being able to expose different features and being able to sort of do things like change image models and really, really dictate some more kind of specific settings around. How wordy do you want it to be? What's your audience? What's the tone, things like that. And so where we are now is you can sort of start very simply with the topic, generates an outline, and then you can choose to sort of dig into the outline, really tweak the settings, and then it generates the whole deck really fast. We actually slow it down. This is really secret Gamma lore. There is this wow moment where people go from the outline to the deck and you can see it in YouTube testing and watch people use it. They're always like, wow, it looks so good. They see it generate and they see it type in and they're just like, there's this moment. They're kind of mesmerized just watching it type in. I think it was maybe like a year or two ago when all the models started getting really, really fast, the text models, and it was like too fast and customers were like, what's happening? And it kept scrolling and just like you sort of started to lose that wow moment, right? And so we slow it down a little bit. But we are going to have an API soon which will, which will give you full speed and not necessarily have to have to go super, super slow, but it's just fast enough for you to like kind of follow along and read it. We really dial it in.
- [21:35]
Rid
You know, it's funny because here I was thinking that maybe we were going to talk about how you're handling all of these different like, loading pages and the time that you have to account for when the AI is generating things and you're totally doing the opposite.
- [21:47]
Zach Leach
Well, well, actually we, we used to have to. In fact, the component is still called like, look over here or something, right? Because it was like, it was like in the code, it was in React. I called it that because there was this like loading screen and all this kind of stuff, like just wait 30, 45 seconds or whatever. Now ducks can generate so fast. When we first built this thing, we were like, okay, so maybe there's something we can give the customer to do. Like maybe they can set some settings. And so we threw the theme picker in there, right? And they were like, it was a full screen kind of moment. You pick your theme and, and do all this stuff. And we got rid of that because we don't need anymore. So finding the right things to like pause on, you know, and finding the things to skip over.
- [22:25]
Rid
It's really neat idea though, pausing on the theme. I get you don't need it anymore. But I'm even thinking in my head, like, I'm working on an AI product right now. There's A couple times where it can sometimes take 10, 12 seconds. And I'm like, I wonder if there's actually a very lightweight task that you can present at that moment as kind of the look over here moment. It's a pretty clever solution, I think.
- [22:44]
Zach Leach
I did some analysis on it, and people would spend up to 30 seconds, like, even. Even after the deck was done, people would spend like 30 seconds just like clicking through all these things and understanding the themes. And also it's an opportunity to introduce the concept of themes, right? Like, oh, we have these. These things and, and sort of here's how they work and here's some that we offer.
- [23:04]
Rid
Anything else that we're not talking about in terms of lessons that you've learned or ways that your thinking has evolved that's now informing this gamma on the principles of designing for AI.
- [23:15]
Zach Leach
Give you more options, give you more variations, I think was something we kind of discovered, you know, especially in some of the image stuff, like being able to kind of give you stuff to pick from. We're working on a feature right now that lets you sort of pick variations of cards, variations of slides, and it just feels fun. I mean, I think ultimately what I ask myself when I'm kind of doing this stuff is like, is this, is this fun to do? Like, is this a fun thing? Or does it feel like work? You know, And I think AI can really make things fun. Another principle that I wrote was encourage rabbit holes or something, right? Where you should be able to kind of follow your creativity and you should be able to kind of go explore and it should feel fast and fluid and you should be able to say, like, I had a paragraph about this. Or like, oh, I had an image about this. Or you're just kind of like moving as fast as you can sort of think it kind of reacts and stuff. And like, you're allowed to go down this rabbit hole and then you can pull back out. Like, that was weird. Like, let me start over or whatever.
- [24:06]
Rid
I remember the first time that someone presented a gamma in our all hands at Maven, and it was Johnny Chen. And he, like, had like kind of the tidy part of the presentation above. And then he scrolled on the presentation and this big full screen goofy gif. And it was kind of, you know, it broke my mental model for what a presentation was in that moment where, like, it wasn't this 16 by 9 box with three bullet points. It was like, whoa, you can kind of get creative here. So I get it. I think you guys are already leaning into that and probably it was a big part of what influenced the latest rebrand as well.
- [24:40]
Zach Leach
Yeah, yeah. Trying to. Trying to get a little fun, a little whimsy, but again, keeping it. Keeping a little buttoned up, I think, too. And then speaking to imagination, a lot of the imagery in the art direction is like this kind of, you know, surrealist sort of like. Like a little bit of a double take on everything. You're kind of like, what is this all about? You know, it's about that. That notion of imagination and kind of like funness and openness and. And then there's an airiness to it, too, I think, which came out really well.
- [25:05]
Rid
My brain typically thinks in spectrums. It's kind of just how I operate. So you have this fun and buttoned up axes. But I would imagine another one that you've probably been thinking a lot about is, you know, is AI more of like a feature or is it the product itself and integrated through every single part? So maybe you could touch a little bit on how you've approached that set of opportunities as a designer and ultimately, like, how it manifested in the interface.
- [25:33]
Zach Leach
Very early on, we didn't have nearly as much conviction about it as we do now. Right. And so, you know, kind of like. I mean, we went for it, but it was also like, is this going to work? You know, and so a lot of what we have in the product is experimental. We'll add in like a. Like an autocomplete thing. We'll add in like a, you know, an image editor. We'll add in like a deck editor. Right. And so I think where we're at right now is sort of. There's a lot of little kind of doors, so to speak, that lead into AI features. And all of them sort of have a little bit of a different vibe, different models and different kind of heuristics a little bit. And one of our big goals is to really try to unify a lot of that. Right. Like. Like, what does it mean? What's the overall vision kind of for. For AI as. As a whole in Gamma. Right. And there's a very simple question of, like, do we say AI? You know, do you just say it or not? Is it just sort of the product? Right.
- [26:25]
Rid
Yeah. You can have 10 different Sparkle icons in so many different levels of the product. And at what point do you even.
- [26:31]
Zach Leach
Yeah, yeah.
- [26:32]
Rid
Indicate that?
- [26:33]
Zach Leach
Yeah, it's sort of like, is, like, at what point is AI going to be just such a part of what we do that you don't need to call it out anymore? I do think that we're going to get to a world which is like, most products are just going to be products and we're not going to say AI, But I also think right now there's a little bit of AI tourism. There's a little bit of people who are kind of looking for that kind of thing as, you know, products evolve and as things change, it's going to be less about, oh, this thing has AI just like, it just does what it does. But then it kind of raises other questions of like, is there one sort of unified kind of, you know, experience to use AI tools in Gamma? Is it, Is it a chat? It's simply just a generator? Is it lots of little kind of contextual help pieces? I don't really know yet, but I can say that our image chat has done. Has done really well. And I think there is something about sort of being able to chat with your whole deck that's very compelling to me. It raises all sorts of other hairy problems, like, how do you know what it can do? Like, what happens when it doesn't do what you thought it was going to do?
- [27:36]
Rid
I want to zoom out from the product for a little bit and talk a bit about how you all work. And I think I want to use a quote from Grant, the CEO, as the entry point because he said if you were the generation before, you'd easily be at 200 employees. So I'm curious, what about the way that you all work allows Gamma to operate at a fraction of the headcount?
- [27:59]
Zach Leach
I think we're a little bit of an unorthodox design team, although maybe this is changing a little bit in sort of the industry. But we're a design team who is quite capable with regards to coding, prototyping, building actual stuff. Like, we close bugs and we ship stuff, not huge stuff, but we can fix things in the product. And more importantly, we can prototype custom builds of the product and share with customers and get feedback on those things. And so as a design team, we're like, pretty capable with sort of that kind of stuff. And, and now that AI has come along, it's really kind of amazing where I can go in and I can have a way better understanding the code base if I'm trying to prototype something out for a customer. Like, recently somebody wanted me to add like an option for analytics in the dashboard. I'm like, let me just see if Cursor can just go do this. Do you know what I mean? Like, like, I seriously think to myself, how many bugs can I as A designer just go fix by just asking Cursor to do it. Like, it's kind of mind blowing.
- [28:54]
Rid
It's a lot. It's a lot. For anyone listening that hasn't done it yet. It's a lot.
- [28:58]
Zach Leach
Yeah, it's like, it's crazy, right? So, but then as far as like the whole company goes, we're really believers, I think here. You know, it's cool being able to see a new model come out and we put it in the product like immediately and just, you know, Deepsea came out and the whole world was like on fire about like this new model or whatever. And like, we tried it out and it turned out like, yeah, it was pretty meh to our customers. Right. Like, ultimately it's, we can put a new model in and then we get feedback immediately about if the deck is good, if the images are good. Like, we can. We have a very sort of strong qualitative kind of measuring quantitative too with regards to like the rating system and everything. And so we're always trying new stuff out. We have great partnerships with a lot of model providers too, and it's always exciting to talk to them and help kind of either guide what they're going to do or get access to something that can really be transformative to what we do here. I can speak to also how midjourney really helped us in our rebrand. We used the style def and personalization stuff to do a lot of our art direct and it's been super helpful. Right, because as a team now we're like, we want an image for like an empty state. Well, we have like a style graph for it. We kind of have, you know, all this stuff and we can kind of do some image generations and not have to kind of go back and forth necessarily with like, does this feel right? And like, you know, who's going to go illustrate this? Like, we can generate stuff that feels very high quality and very on brand and sort of very consistent because of some of the tools that midjourney has.
- [30:22]
Rid
So I want to go down the like, code prototyping rabbit hole. But really quickly, you mentioned something about the rating system. Can you explain what that is for people?
- [30:31]
Zach Leach
Yeah. So every time you generate a deck at the bottom, you basically get a sad face, a meh face, you know, and then a smile, a happy face. And it's just like very lightweight, very low touch, little, little way to just see like, you know, was this good or was this bad? And. And what we can do is basically a B test models. We can put a model out to 10%, whatever, 50%, and then we have enough volume that we can just basically say, was this getting more smiley faces or not? For this given model, we have kind of a leaderboard internally, like, oh, this thing is the new top. The top dog, right?
- [31:07]
Rid
Oh, very cool.
- [31:08]
Zach Leach
We're not Google Slides. We're not stuck on Google model. We're not PowerPoints. We're not stuck on the Microsoft models. We kind of pick and choose. We're sort of in a great spot to just literally figure out what's the best model to deliver to our, to our customers, what's the best model for generation, and just do that. We're not, we're not. Our hands aren't tied to that capacity, so.
- [31:30]
Rid
Power to the app layer.
- [31:31]
Zach Leach
Yeah. So if you look at like, you know, PowerPoint's gonna always have to use the Copilot model, you know, and Google's always going to use Gemini. And like, we can be like, well, actually the outline works best with this model, and then the generation works best with this model and the images work best with this model. So. So, like, we're just, we're just out here trying to give people the best, best stuff possible. And the ratings work for decks and they work for, for images. And so we have sort of a little bit of a competition going on internally. But, yeah, it makes sense.
- [31:58]
Rid
I mean, this is something that I'm kind of noticing in a lot of these interviews with people who are designing AI products is there's always some mechanism to capture, like, a quantitative reaction to any of the AI output. And even for myself, I'm like, listening to you talk, I'm like, huh, okay. How can I introduce something like that?
- [32:16]
Zach Leach
Well, also, we all come from Optimizely, so we're sort of like, we're like very into kind of like measuring stuff and making sure that we do really care about. We do a tremendous amount of data collection and as far as, like, analytics and the product and like, what, you know, you can look at what's working well, what's not working well. And designers too are empowered to kind of do some of that analysis, you know, and go understand where are things, where are things falling down and like, what sort of some of the problem areas.
- [32:42]
John
I'm a big believer in the power of video to explain my thinking as a designer. So when it's time to get feedback, I'll drop a loom link and slack and another link to a Figma prototype.
- [32:52]
Rid
And feedback will be scattered everywhere.
- [32:54]
John
And I mean, it's a mess. So I'm building the product that I've always wanted to exist and it's called Inflight. You can kind of think of it like an Async crit. It's an easy way to share a video walkthrough along with an interactive prototype or whatever you're designing. And then AI interviews the people on your team to get you the feedback that you need and organizes everything for you in a beautiful insights page. So right now I'm only giving access to Dive Club listeners.
- [33:21]
Rid
So if you want to be one.
- [33:22]
John
Of the first to use Inflight, head to dive, dot club, slash inflight to claim your spot.
- [33:28]
Rid
I want to talk about the technical piece now because from our previous conversation, it's not just you too. Like, a lot of the designers at Gamma are more technically minded, comfortable getting into the code.
- [33:39]
Zach Leach
Yeah.
- [33:40]
Rid
And maybe even just as like a quick context piece. Like how much of a technical background did you have before Cursor and this whole AI wave?
- [33:48]
Zach Leach
I learned React at Gamma. I was just reminded that actually recently, like, hey, Zach, remember when you didn't know React? I was like, oh yeah. But you know, I've always had a little bit of technical background. You come in and you're kind of like, well, I can kind of like write HTML, I kind of know CSS and kind of get by. But as soon as you're like, let me just ask ChatGPT about this thing, like if you have any kind of, you know, desire to want to learn and to grow and to like understand this stuff, like the tools are out there now, right. You can get in to ChatGPT or Cloud or whatever and just, and just ask questions. And then more importantly, operating on the real code base has been a real game changer. Right. Because you can just ask questions, not in the abstract, but like, how does this button work? Or what does this do? Or like, why isn't this loading? Right. We also have a lot of design engineering support too. Right. So we have design engineer who helps us do a lot of prototyping and can work through that stuff. And, and ultimately it's about being able to deliver the vision to the engineers in a really clear, put together way that feels real, but also being able to test that with customers and being able to make a prototype that's testable. It's really hard to sort of imagine what a FIGMA prototype of this thing would be. You can make in Figma, like very specific screens and very specific flows, but when you put it in the hands of a customer who can make anything they want. They can make as many cards as they want. They can type whatever they want. They can do, you know, all of this different stuff. The prototype becomes so much more real and so much more. More alive, and you can really see where things break or. Or work really well when you kind of get in front of people in a real working way. We. We use a ton of. Of code Sandbox and Stack Blitz. I've played with both a lot, too. Just trying to, like, make stuff and make little prototypes, like, for some of the AI generator stuff. I just made, like, a Bolt prototype, and I'm like, let me just do this, you know, Let me just see what this would feel like. I wish Bolt had the ability to have some notion of our design system. I think that'd be really cool, because right now, it doesn't know about our company. Kind of speaks to our. That last thing we're talking about, which is, like, if it knew more about who you were, knew more about your company, knew more about what you're trying to do, like, I think that would be even. Even cooler. But we'll see if that happens.
- [35:56]
Rid
All right, so we talked about, like, the technical piece. I do kind of want to even return to the beginning of this conversation where, you know, you've been working with this group of people for quite some time now. There's a familiarity and probably a really healthy collaboration that you've ironed out and refined over years. Are there other ways that that way of working has evolved now that you're dealing directly with AI as a material?
- [36:20]
Zach Leach
The Gamma's four years old, and we spent six years at Optimizer, so that's like a decade of. Of working with, like, some of the same people, which is incredibly powerful. I feel like a lot of times there's just such so much trust and so much kind of faith and confidence in each other. It's amazing. I think that that really actually came to a head when we all decided to do this AI thing, right? I mean, you got to put yourself back in that kind of era where, again, there was no chatgpt. There was nothing. And the models were slow and janky and, like, could barely spell stuff correctly. And we're like, you know what we're gonna do? We're gonna make an AI model that generates our slide decks. And it took a lot of faith to want to do that, to want to say, like, I believe in this thing. And, like, we all kind of had to trust each other to want to do that. And, you know, we trusted each other to make the Prototypes. We trusted each other to kind of make it happen and it was, it turned out pretty good.
- [37:16]
Rid
What about in your personal process? Like, I'm sure you're still doing a lot of kind of open ended exploration in figma. It sounds like you're actually, you know, punching out tickets and shipping fixes for bugs and, you know, you're making these functional prototypes in Bolt and Cursor. Like, how do you even allocate what tools and the medium in which you are working based off of like the project or where you're at in the design process? Because I'm sure there's somebody listening and they're like, yes, you know, cursor is so empowering. But like, when do I use it? You know, like, when does an actual professional designer working at an AI company take advantage of these tools versus work a little bit more traditionally?
- [38:00]
Zach Leach
I think you could do this. I think if there's traditional designers out there and people who are making figmas and stuff, just see what happens. Take your figma, put it in Bolt and say, make a prototype. Like, I bet you get pretty far. Like, you know what I mean? It's pretty good. Now it's not going to be perfect, right? But then you ask yourself, what do you want to learn, right? What are you trying to learn here? Is this something I could put in bold, spend 20 minutes kind of refining and then go send it to a customer really quick, like, probably, right? Like, are they gonna care? Are they gonna care? It doesn't look perfect or like work perfect. But, but if, if the question you're trying to answer is like, you know, does this work? Can this work? Like, where does this fall down? You can do an enigma, right? But then you could also, you know, take something like Bolter or lovable or whatever, make something that feels a little bit more real and try it for yourself. If you know it really works, just share it around. Post it on Slack. Like say, hey, I was just experimenting with this. Here's like a working prototype of this thing kind of working. What do you guys think?
- [38:55]
Rid
Is there an example we could use of a time where you had some kind of a learning goal or something? You were trying to figure out where you reach for one of these tools. And then we could talk a little bit about like the outcome and what it looked like to iterate from that point.
- [39:06]
Zach Leach
We can make a deck and the decks have images in them, but a lot of times the image is like, oh, it's a little bit off, or like, oh, this table's weird. Or there's missing fingers or whatever. So we wanted people to be able to go in and say, tweak this image, make this image a little bit better. What's wrong with it? Let me tell you. Here's kind of what we're going to change. I can like work all this stuff out and try to like do some Figma stuff and talk some customers or I could just like have Bolt just like make something, you know, just like to see what happens. And I did make an entire prototype of this AI image flow. I actually made three of them in. It kind of iterated on all of them and it ended up, you know, not really using a whole lot of it, but it was almost just a tool to kind of just like explore. And what I'm looking for is, is it fun, right? Like, was this thing fun to use? And I could really find the fun easier with something like Bolton just telling it like, oh, actually like give me four variations, don't give me one anymore, or always make one a little bit more, you know, vibrant or like whatever it might be during the image generating experience. And it was just easier for me to kind of like find that fun. And ultimately what happened with it was I sort of like, you know, it gets to sort of his logical conclusion, like, can't take this thing any further. And it's kind of answer the questions you wanted to answer. And then you, you use that to inform kind of another design iteration process. Right. Maybe you do it for real in the product or maybe you do it again in Figma and share it around the team and crit it and stuff.
- [40:25]
Rid
I love find the fun.
- [40:26]
Zach Leach
Yeah, I kind of do find the fun in a presentation tool. Yes.
- [40:31]
Rid
It's something that I've even done because I use, use lovable a lot, you know, kind of accomplishing similar things. And I found a lot of success in having like ready made starting points that I then duplicate before I start adding any code or like working on a specific prototype. So I kind of have like this shell of the product that you're right. Like it's probably like 80, 90% fidelity, but it's totally good enough to kind of get what it feels like. And, and then you just build a little piece on top of it, like, okay, then I duplicate it again, build a little piece on top of it and it's, it's become a pretty stable component of my practice now.
- [41:06]
Zach Leach
Yeah, yeah. And I think Bolt just last week introduced like some new design update thing and it's actually really good. I was like, dang, I like I was like, just messing with it. I made like a recipe app or something and like it made like like a whole color scheme, like seraphims, like really nice hours. I was like, wow, this is like very decent, you know, I mean it's like sort of democratized, but also like very like, oh, cool. I like put it together pretty nice.
- [41:27]
Rid
You've mentioned co generation and then also using Mid Journey for like empty states, things like that. Any other applications for how you're using AI in your practice?
- [41:37]
Zach Leach
We have a shared claw project that we use for like helping us write job descriptions, which is pretty cool. As we're hiring, we have like a bunch of example ones and we're like, oh, kind of like start me out with something for, you know, this job or whatever that's been cool to use. And it's like such a simple sort of concept where you just like put a bunch of them in there and say like, generate another one and send like a shared project. Like it's no, by no means is like some fancy tool or anything. It's just like we're all in the cloud project together. If you can take some topic that you want to like share or really learn about. I've been using OpenAI's deep research and then piping that into Gamma and it is, it's great. It's very thorough and it keeps the references and I wish this was like a thing, you know, Like, I wish, I wish Gamma could do this somehow because some of the outputs of this is just like so fascinating.
- [42:27]
Rid
So you're just using the API that is not public yet, right?
- [42:30]
Zach Leach
No, no, no. I totally go to Chat GPT and do a Deep Research and then copy and paste it into Gamma like this.
- [42:35]
Rid
Copying and pasting it in. So it's just a way to visualize deep research.
- [42:39]
Zach Leach
Yeah, but like, especially if you do like web image searches, it'll like go find the sources and stuff. It's. Yeah, it's very cool. I might prototype something around this because I really, really think it's cool. Very, very, very nitty gritty stuff. I use replicate some of their APIs for like image manipulation, upscaling and background removal and stuff like that.
- [43:00]
Rid
Yeah, it's cool. I mean, just seeing the spectrum of all the different things that you're getting your hands into, from marketing to much more technical functions. And maybe before I let you go, we can kind of zoom in all the way out because you're really in the weeds in terms of what AI is capable of today. But if you kind of extrapolate it in your mind a little bit. How does it shift the way that you think about your role as a designer in the coming years?
- [43:26]
Zach Leach
If our goal is to make great tools that are, that are efficient and fun to use, like, I honestly think we're going to be able to make prototypes faster. I really hope in the future there's going to be like better AI tools around user research and analysis. We have a tremendous amount of data, we do a lot of analytics and what's working, what's not working. Like, it'd be so cool to be able to use AI in that space too and understand some customer sentiment more deeply. I think as things evolve and as those tools become reality, a designer is going to be able to not write a metabase query, but go ask metabase like what's been going on, where are some problem areas and then go right to something like Bolt or whatever and be able to prototype something out for that. And it knows about your design system, it knows what you've been doing. I think we're just going to be more empowered and be able to make better decisions. And the Gamma design team ethos is all about being able to deliver their vision in a high fidelity way in a way that really makes sense. I think we're only going to be able to do that better as these tools get better and as we can work with better information.
- [44:31]
Rid
Kind of just looking ahead into the Gamma future, is there anything that you can share about what you're kind of imagining or excited about for where this product can go?
- [44:40]
Zach Leach
You know, we found great success in, in presentations and with this medium. But I think the future is going to be how do we take that, those learnings and how we take like the goodness of that AI and, and all the, the ability to make stuff beautiful by default and apply it to other stuff. Right. Like we're talking websites, we're talking maybe docs or maybe even, you know, video, audio, whatever it might be. We have sort of hit our stride here for presentations and it's going to be very interesting to see like, like where that goes with other mediums and other, other formats.
- [45:08]
Rid
I should say, Zach, rooting for you, big fan of everything that you're doing and I appreciate you taking the time and pulling the curtain today and just kind of sharing a little bit about how you all work and also just like the specific challenges that you're overcoming or ways that you're kind of wrangling the AI, it's definitely shedding a lot of light on what it's like to work on this kind of a product.
- [45:26]
Zach Leach
Yeah. Thank you. Thanks. I appreciate it.
- [45:28]
John
Before I let you go, I want to take just one minute to run you through my favorite products because I'm constantly asked what's in my stack. Framer is how I build websites. Genway is how I do research. Granola is how I take notes during crit. Jitter is how I animate my designs. Lovable is how I build my ideas in code. Mobin is how I find design inspiration. Paper is how I design like a creative. And Raycast is my shortcut every step of the way. Now, I've hand selected these companies so that I can do these episodes full time. So by far the number one way to support the show is to check them out. You can find the full list at Dive Dot Club Partners.